Developing a Probabilistic Heavy-Rainfall Guidance Forecast Model for Great Lakes Cities

Abstract

A method for predicting the probability of exceeding specific warm-season (April-October) 0-24 hour precipitation thresholds is developed based upon daily maximums of meteorological parameters. North American Regional Reanalysis and Daily Unified Precipitation data from 2002-2017 were used to gather meteorological data for the Milwaukee and Chicago County Warning Areas. Individual artificial neural networks and multiple logistic regressions were conducted for daily rainfall thresholds above 0.5'', 1'', 1.5'' and 2'' to determine the probability of threshold exceedances for each County Warning Area. The most important parameters were 1000-500 hPa specific humidity, vertical velocities at various levels, high cloud cover, precipitable water percentile relative to climatology, and surface convergence. Critical Success Indices were universally higher than the average 2017 warm-season WPC threat scores across all thresholds, showing potential promise in operational forecasting use. Sensitivity analyses were conducted to determine degradation of model results when using NWP model forecasts, with mixed results between the two cases studied. Future work includes using additional years of reanalysis and rainfall data to increase heavy-rainfall case counts and boost model skill, as well as to include additional case studies to further analyze model degradation when using NWP model forecasts.